Spectral Screening Based on Comprehensive Similarity and Support Vector Machine

• Published in: IEEE journal of selected topics in applied earth observations and remote sensing 2019-12, Vol.12 (12), p.4854-4860
• Main Authors: Zhan, Yunjun, Dai, Tengda, Su, Yubin, Huang, Jiejun, Liu, Congyuan
• Format: Article
• Language: English
• Subjects: Computation; Distance measurement; Hyperspectral imaging; Mapping; Minerals; Remote sensing; Screening; Similarity; Spectra; Spectral analysis; Spectrum analysis; Support vector machines; support vector machines (SVMs)
• ISSN: 1939-1404; 2151-1535
• DOI: 10.1109/JSTARS.2019.2925906

Using field spectra for hyperspectral remote sensing minerals mapping has good applicability. However, the spectra collected in the field must be screened before applying to hyperspectral mapping, and it is difficult to screen the field spectra quickly and efficiently. In order to rapidly and efficiently screen spectra, we propose a new screening method. In this method, we construct a comprehensive similarity evaluation model, in which local similarity and global similarity are considered synthetically, and then combined to support vector machine, which is used to set the threshold, to screen the field spectra. The method can prevent the result of computation from trapping in local or overall similarities during spectral matching, and can objectively and automatically set the threshold value for eliminating abnormal spectra. According to our experimental verifications, this method is feasible and efficient for eliminating abnormal spectral, thus it can efficiently improve recognition accuracy. This spectral analysis method can also be used in other fields for abnormal spectral elimination.